New research aims to make AI better at understanding important information in language-based technologies

By Line Nyrup Odgaard, AAU Communication and Public Affairs
Photo: Xinyuan Liu

Marcell Richard Fekete from the Department of Computer Science has received DKK 2.1 million from Independent Research Fund Denmark for a research project running from 2026 to 2028. The project investigates how artificial intelligence can become better at understanding what is important in a text or conversation, particularly when language is combined with other types of information such as images or audio.

When people communicate, we naturally highlight what is new or most relevant, but since AI technologies are not typically evaluated on this ability, we cannot know whether they do the same. This is why Marcell Fekete is developing a new dataset that shows how humans typically organize information. The dataset will be used to test how well current AI models recognize these patterns and where their limitations lie.

The project focuses particularly on language‑based technologies such as voice assistants, live captioning, and automatic transcription, which are increasingly used by people with disabilities to communicate and access information.

At the same time, the project aims to generate new knowledge about how AI models understand and organize information, so these technologies can become more reliable, accurate, and user‑friendly — benefiting the many people who rely on digital solutions in their daily lives.